docs(examples): curate to 47 scenario-shaped examples (#37)

* docs(examples): curate to 47 scenario-shaped examples

Cuts the example set from 102 to 47 and reshapes every keeper around a
recognizable real-world situation a developer would say "yes, that's
my problem" to — replacing the API-tour shape that read like an
integration test.

# What every keeper looks like now

Standard header at the top of every file:

  //! What you'll learn:    <one-sentence takeaway>
  //! Why this matters:     <when you'd reach for this in your code>
  //! Scenario:              <the situation the body enacts>
  //! Run with:              <exact cargo command>
  //! Sample output:         <real output captured against ollama / llama3.1>

So a developer browsing the repo can decide whether the example is what
they want without running it.

# Scenarios (a sample)

- `memory_types` — a user introduces themselves, plans a trip, then
  asks the agent to recall their name from turn 1; we run the same
  conversation through Buffer / Window(2) / TokenBufferMemory(200)
  and watch which still remember.
- `agents_conversational` — multi-turn recall test ("what's my name").
- `agents_multi_agent` — Sequential planner -> executor for a real
  team-standup task.
- `chains_structured_extraction` — parse a meeting note into a typed
  Vec<ActionItem { who, what, due }>.
- `chains_pipe_operator` — triage incoming support tickets through
  embed -> categorize -> tag.
- `models_embedding` — find the most similar product description for
  a search query.
- `models_routing` — route greetings to a fast model, technical
  questions to a bigger one.
- `tools_orchestrator` — fan out three slow price-fetch tools in
  parallel, fold to the cheapest.
- `retrieval_indexing_rag` — docs site re-indexer; round 1 indexes 3
  docs, edit one, round 2 only re-embeds the changed doc.
- `graphs_with_checkpoints` — resumable 5-step ETL pipeline (extract /
  validate / transform / load / notify) with per-step snapshots.
- `graphs_interrupts` — agent drafts a customer email; graph pauses
  before the `send` node so a reviewer can approve, then resumes.
- `obs_evaluation` — regression eval over 5 known FAQ pairs you'd run
  in CI to catch quality regressions.
- `resilience_ssrf_protection` — URL-fetch tool that gets a URL from
  user input; reject 10.x / 192.168.x / localhost / 169.254 (cloud
  metadata) before the request leaves your process.

# Cuts

Removed examples that were:
- staged demos using fake providers (StaticEcho / EchoProvider /
  CannedProvider / FakeProvider / Tagged)
- hand-crafted AiMessage / Message construction to demonstrate
  internals
- partial-step tours of an API surface that other examples already
  cover end-to-end (chains_simple, chains_composition, retrieval/{
  vector_store, document_store, vector_store_search,
  rag_with_vectorstore, document_compression, cross_encoder,
  document_transformers}, graphs/{validation, visualization, runtime,
  metrics, debug, dot_export, analysis, ...})
- pure-internal demos (obs_callback_manager, obs_tracing,
  agents_bus_pubsub, agents_lifecycle, agents_execution_hooks,
  agents_plugin_system, agents_planning_middleware)

55 files removed, 7 merged, 47 retained.

# Verification

- `cargo build -p cognis-examples` clean.
- All 47 ran end-to-end against Ollama (llama3.1 + nomic-embed-text);
  captured sample output is in each file's header.
- `graphs_interrupts` rebuilt to compile two graph variants (one with
  interrupts, one without) so resume actually advances past the
  pause point.

# Docs

- `docs/mintlify/examples/*` — 12 mdx pages updated to reflect the
  curated list. Tables now read "Name | Scenario | Source".
- `docs/mintlify/examples/index.mdx` — counts updated, header
  template documented for first-time visitors.
- `README.md` — example commands list refreshed.

# Out of scope

- `crates/v1/` — frozen V1 source, untouched.
- `models_content_blocks` — needs a vision model (`llava` or hosted
  vision-capable model). Captured output reflects llama3.1 (no vision)
  refusing the request; header notes the model requirement.

* chore: apply rustfmt to all 47 examples

Pre-push checklist miss on the previous commit. Running `cargo fmt --all`
on the curated example set; no behavior changes.

Author: 0xvasanth

README.md

@@ -276,28 +276,26 @@ cognis-rag = { version = "0.2", features = ["faiss", "openai"] }
 git clone https://github.com/0xvasanth/cognis.git
 cd cognis
 
-# Offline demos (no API keys needed)
-cargo run -p cognis-examples --example chains_pipe_operator
+# Offline demos (no LLM needed)
 cargo run -p cognis-examples --example tools_orchestrator
-cargo run -p cognis-examples --example agents_round_robin
-cargo run -p cognis-examples --example agents_bus_pubsub
-cargo run -p cognis-examples --example memory_entity
-cargo run -p cognis-examples --example memory_knowledge_graph
-cargo run -p cognis-examples --example retrieval_document_transformers
 cargo run -p cognis-examples --example graphs_state_machine
-cargo run -p cognis-examples --example graphs_dot_export
-cargo run -p cognis-examples --example resilience_advanced_rate_limiters
-cargo run -p cognis-examples --example parsers_fixing
-cargo run -p cognis-examples --example parsers_retry
-
-# Provider-backed demos (need a running LLM)
-COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.2:1b \
-  cargo run -p cognis-examples --example agents_react_agent
-COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.2:1b \
+cargo run -p cognis-examples --example graphs_with_checkpoints
+cargo run -p cognis-examples --example graphs_interrupts
+cargo run -p cognis-examples --example resilience_ssrf_protection
+cargo run -p cognis-examples --example resilience_rate_limiters
+
+# Provider-backed demos (need a running LLM — `ollama pull llama3.1`)
+COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.1 \
+  cargo run -p cognis-examples --example 02_five_line_agent
+COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.1 \
+  cargo run -p cognis-examples --example 06_ollama_tool_calling
+COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.1 \
+  cargo run -p cognis-examples --example agents_conversational
+COGNIS_PROVIDER=ollama COGNIS_OLLAMA_MODEL=llama3.1 \
   cargo run -p cognis-examples --example retrieval_rag_pipeline
 ```
 
-The full demo set lives under [`examples/`](examples/) — organized by `chains/`, `agents/`, `memory/`, `models/`, `tools/`, `retrieval/`, `graphs/`, `observability/`, `resilience/`, and `parsers/`. Every example is registered in [`crates/examples/Cargo.toml`](crates/examples/Cargo.toml).
+47 curated examples live under [`examples/`](examples/) — organized by `v2/` (the numbered onboarding tour), `chains/`, `agents/`, `memory/`, `models/`, `tools/`, `retrieval/`, `graphs/`, `observability/`, `resilience/`, and `parsers/`. Every example file opens with a `What you'll learn / Why this matters / Scenario / Run with / Sample output` header so you can decide whether it's what you want before running.
 
 ## Build & test